Narrow Results

This study attempts to forecast Bitcoin using both traditional and machine learning approaches to determine which methods are more robust. For the traditional method, the GARCH method is used to forecast Bitcoin returns. For the machine learning method, LSTM is used. A hybrid approach of GARCH–LSTM is also applied to the data to compare the results. Hourly data for Bitcoin are obtained from Coin-MarketCap for three years, from 2019 to 2022. The findings of the study reveal that machine learning methods outperform traditional methods. The study has useful implications for researchers.

This chapter presents Algorithmic Analyst (ALAN), an application that implements statistics and artificial intelligence methods with natural language generation to publish multimedia financial reports in Chinese and English. ALAN is a portion of a long-term project to develop an Artificial Intelligence Content as a Service (AICaaS) platform. ALAN gathers global capital market data, performs big data analysis driven by algorithms, and makes market forecasts. ALAN uses a multi-factor risk model to identify equity risk factors and ranks stocks based on a set of over 150 financial market variables. For each instrument analyzed, ALAN computes and produces narrative metadata to describe its historical trends, forecast results, and any causal relationship with global macroeconomic variables. ALAN generates English and Chinese text commentaries in html and pdf formats, audio in mp3 format, and video in mp4 format for the US and Taiwanese equity markets on a daily basis.

After admitted by the intensive care unit (ICU), a patient may experience mechanical ventilation (MV) if he/she suffers from acute respiratory failure. Vital signs and lab tests associated with the patient are typically recorded in a series over time. We propose an LSTM-based deep relative risk model to quantify patients’ time to occurrence of MV. The internal time-varying covariates motivate us to learn the ratio function via an LSTM net. The number of LSTM cells equals to the width of the sampling window; that is, the i-th cell of the LSTM net takes the patient’s covariates of the time interval i as an input. A subsequent linear layer is used to summarize the hidden layers as the final partial likelihood contribution of each individual. Such an architecture solves the survival analysis problem with internal time-dependent covariates in a nonparametric way. Our experiments based on the MIMIC-III database demonstrate it is a very promising approach to predicting the occurrence of MV.

Assessing the impacts of climate change on hydrological systems requires accurate downscaled climate projections. In the past two decades, various statistical and machine-learning techniques have been developed and tested for climate downscaling; however, there is no consensus regarding which technique is the most reliable for climate downscaling and hydrological impact assessment. In this study, an advanced machine-learning technique, Long Short-Term Memory (LSTM) neural network, is used to build multi-model ensembles for downscaling climate projections from a wide ranges of global and regional climate models, and its performance is compared with a number of traditional statistical and machine-learning methods, such as ensemble average, linear regression, Multi-layer Perceptron, Time-lagged Feed-forward Neural Network, and Nonlinear Auto-regression Network with exogenous inputs. The downscaling input consists of temperature and precipitation projections provided by regional climate models, such as CanRCM4, CRCM5, RCA4, and HIRHAM5, and the output is observation data collected from meteorological stations. Performance of the developed LSTM ensemble is evaluated for two case studies in Canada and China. The downscaled climate projections are further used to assess the hydrological impacts in the southwestern mountainous area in China, with the assist of a fully distributed hydrological model, MIKE SHE. The results can support future applications of LSTM neural networks and other similar data-driven techniques for climate downscaling and hydrological impact assessment.

Artificial neural network (ANN) is a prevalent tool because of its extensive adaptivity and outstanding performance. According to previous studies, Long Short-Term Memory (LSTM) neural networks generally perform well in forecasting financial time series than other models. However, few studies apply LSTM to CPI and price level forecasting. This paper separately constructs the LSTM and the Vector Autoregression (VAR) model, a classic econometric approach for time series forecasting, based on 23 factors that affect CPI directly or indirectly. The results show that the error of the LSTM is significantly lower than that of the VAR in forecasting China’s CPI, while the VAR model provides an explicit explanation of the factors of CPI forecasting through the Granger causality test. Additionally, a synthetic model combining the advantages of both generates a more satisfying outcome. This paper forecasts the CPI by combining the LSTM and VAR models for the first time and provides a new reference to the inflation forecasting area.

Protein domain boundary prediction is usually an early step to understand protein function and structure. Most of the current computational domain boundary prediction methods suffer from low accuracy and limitation in handling multi-domain types, or even cannot be applied on certain targets such as proteins with discontinuous domain. We developed an ab-initio protein domain predictor using a stacked bidirectional LSTM model in deep learning. Our model is trained by a large amount of protein sequences without using feature engineering such as sequence profiles. Hence, the predictions using our method is much faster than others, and the trained model can be applied to any type of target proteins without constraint. We evaluated DeepDom by a 10-fold cross validation and also by applying it on targets in different categories from CASP 8 and CASP 9. The comparison with other methods has shown that DeepDom outperforms most of the current ab-initio methods and even achieves better results than the top-level template-based method in certain cases. The code of DeepDom and the test data we used in CASP 8, 9 can be accessed through GitHub at https://github.com/yuexujiang/DeepDom.